Do Children with Dyslexia And/Or Specific Language Impairment Compensate for Place Assimilation? Insight Into Phonological Grammar and Representations Chloe R

This article was downloaded by: [Harvard College] On: 25 October 2011, At: 03:18 Publisher: Psychology Press Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Cognitive Neuropsychology Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/pcgn20 Do children with dyslexia and/or specific language impairment compensate for place assimilation? Insight into phonological grammar and representations Chloe R. Marshall a , Franck Ramus b & Heather van der Lely b c a Department of Language and Communication Science, City University London, London, UK b Département d'Etudes Cognitives, Ecole Normale Supérieure, Paris, France c Department of Psychology, Harvard University, Cambridge, MA, USA Available online: 30 Jun 2011

To cite this article: Chloe R. Marshall, Franck Ramus & Heather van der Lely (2011): Do children with dyslexia and/or specific language impairment compensate for place assimilation? Insight into phonological grammar and representations, Cognitive Neuropsychology, 27:7, 563-586 To link to this article: http://dx.doi.org/10.1080/02643294.2011.588693

PLEASE SCROLL DOWN FOR ARTICLE

Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7), 563–586

Do children with dyslexia and/or speciﬁc language impairment compensate for place assimilation? Insight into phonological grammar and representations

Chloe R. Marshall1, Franck Ramus2, and Heather van der Lely2,3 1Department of Language and Communication Science, City University London, London, UK 2De´partement d’Etudes Cognitives, Ecole Normale Supe´rieure, Paris, France 3Department of Psychology, Harvard University, Cambridge, MA, USA

English speakers have to recognize, for example, that te[m] in te[m] pens is a form of ten, despite place assimilation of the nasal consonant. Children with dyslexia and specific language impairment (SLI) are commonly proposed to have a phonological deficit, and we investigate whether that deficit extends to place assimilation, as a way of probing phonological representations and phonological grammar. Children with SLI plus dyslexia, SLI only, and dyslexia only listened to sentences containing a target word in different assimilatory contexts—viable, unviable, and no change—and pressed a button to report hearing the target. The dyslexia-only group did not differ from age-matched controls, but the SLI groups showed more limited ability to accurately identify words within sentences. Once this factor was taken into account, the groups did not differ in their ability to compensate for assimilation. The results add to a growing body of evidence that phonological representations are not necessarily impaired in dyslexia. SLI children’s results suggest that they too are sensitive to this aspect of phonological grammar, but are more liberal in their acceptance of alternative phonological forms of words. Furthermore, these children’s ability to reject alternative phonological forms seems to be primarily limited by their vocabulary size and phonological awareness abilities.

Keywords: Phonological representations; Place assimilation; Dyslexia; Speciﬁc language impairment.

Signiﬁcant aetiological overlap has been reported comorbidity of two in particular, dyslexia and between many developmental disorders, and the speciﬁc language impairment (SLI), is receiving Downloaded by [Harvard College] at 03:18 25 October 2011

Correspondence should be addressed to Heather van der Lely, Department of Psychology, Harvard University, Cambridge, MA, USA. (Email: [email protected]) We are very grateful to Mike Coleman for his programming wizardry. We thank Anastasia Archonti, Hillit Dayan, Sally Harcourt-Brown, Angela Pozzuto, and Sophie Tang for their assistance in testing participants. We are most grateful to all the children who participated in this project, as well as their schools and their parents. We are also grateful to the feedback from various audiences to whom we have presented this work, at Reading University, the University of Amsterdam, the Laboratoire des Sciences Cognitives et Psycholinguistique in Paris, and the 16th Manchester Phonology Meeting. The editor, Brenda Rapp, and two anonymous reviewers provided extremely helpful comments on previous versions of the paper. This research was conducted at the Centre for Developmental Language Disorders and Cognitive Neuroscience, London. The ﬁrst author and the research were supported by an Economic and Social Research Council (ESRC) grant awarded to H.v.d.L. (RES–000–23–0575).

# 2011 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business 563 http://www.psypress.com/cogneuropsychology DOI:10.1080/02643294.2011.588693 MARSHALL, RAMUS, VAN DER LELY

much attention in the current literature (Bishop Lu, Manis, & Seidenberg, 2005), a perceptual- & Snowling, 2004; Catts, Adlof, Hogan, & centres perception deficit (Goswami et al., 2002), Weismer, 2005; Pennington & Bishop, 2009; an anchoring deficit (Ahissar, 2007), and abnormal van der Lely & Marshall 2010; inter alia). temporal sampling (Goswami, 2011). Developmental dyslexia is defined as an impair- Nonetheless, researchers generally agree that a ment in acquiring literacy skills despite normal phonological deficit is one of the most prominent sensory abilities and nonverbal IQ and adequate symptoms of dyslexia and attempt to explain it one exposure to written language (Snowling, 2000). way or another. While it is widely acknowledged SLI is defined as an impairment in acquiring that there is a reciprocal relationship between language despite, again, normal sensory abilities reading and phonological skills so that the phono- and nonverbal IQ and adequate exposure1 logical deficit theory is partly circular (Castles & (Leonard, 1998). The overlap between the two dis- Coltheart, 2004), there is nevertheless ample orders is substantial, with over half of dyslexic chil- longitudinal evidence for the precedence of pho- dren having SLI and vice versa (McArthur, nological and more general language-related defi- Hogben, Edwards, Heath, & Mengler, 2000). cits in children at risk of becoming dyslexic Several theories of dyslexia and SLI claim that pho- (Guttorm, Leppa¨nen, Richardson, & Lyytinen, nological deficits of various types underlie both dis- 2001; Lyytinen et al., 2004; Molfese, 2000; orders and therefore play a key role in this overlap Scarborough, 1990). Investigating the nature and (Joanisse, Manis, Keating, & Seidenberg, 2000; the cause of the phonological deficit of dyslexic Kamhi & Catts, 1986; Messaoud-Galusi & children thus remains a major research goal. Marshall, 2010; Tallal, 2003). The phonological deficit in dyslexia makes itself There is considerable evidence for a phonological manifest in three main areas: manipulating phono- deficit in dyslexia (for some early work, see Bradley & logical representations (e.g., phoneme deletion Bryant, 1983; Lundberg & Høien, 1989; Wagner & tasks), holding verbal material in short-term Torgesen, 1987), although it has been argued that a memory (e.g., nonword repetition and digit span phonological deficit is neither necessary nor suffi- tasks), and accessing phonological representations cient to cause dyslexia and that the deficit might be (rapid naming tasks). What underlies this phono- much broader. Alternative theories of dyslexia logical deficit is less clear—degraded (i.e., fuzzier, include theories that ascribe the reading disability noisier, or underspecified) or, conversely, overspe- directly to purely visual or visual attention deficits cified phonological representations, limited (Stein & Fowler, 1981; Valdois, Bosse, & working memory capacity, and speech perception Tainturier, 2004; Vidyasagar & Pammer, 2010). problems have all been proposed and might be Other hypotheses focus on underlying cognitive interrelated (Adlard & Hazan, 1998; Mody, and/or neural causes of either the phonological Studdert-Kennedy, & Brady, 1997; Serniclaes,

Downloaded by [Harvard College] at 03:18 25 October 2011 deficit (such as auditory theories, Tallal, Merzenich, Van Heghe, Mousty, Carre, & Sprenger- Miller, & Jenkins, 1998), or of broader manifes- Charolles, 2004; Snowling, 2000). A major tations including phonological, visual, and learn- current research question concerns whether the ing/memory deficits (such as the magnocellular phonological deficit consists of an actual degra- theory, Stein & Walsh, 1997; the automaticity/ dation of phonological representations themselves, cerebellar/procedural learning theory, Nicolson & or whether it is a deficit in accessing and manipu- Fawcett, 2007; the sluggish attentional shifting lating those representations (Blomert, Mitterer, & theory, Hari & Renvall, 2001). Recent theoretical Paffen, 2004; Dickie, 2008; Ramus & Szenkovits, proposals include a noise exclusion deficit (Sperling, 2008; Soroli, Szenkovits, & Ramus, 2010).

1SLI also occurs in signed languages, among deaf children who are acquiring a signed language as their native language (Mason et al., 2010).

564 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

The picture with regard to SLI is even more characteristic of particular sound sequences rather muddled. Although children with SLI are than particular words. One such pattern involves reported to have difficulties with phonological rep- a process where a particular type of sound is sys- resentations, phonological working memory, and tematically altered when it is juxtaposed with speech perception (Bortolini & Leonard, 2000; another particular type of sound. For example, in Gallon, Harris, & van der Lely, 2007; Leonard, English this situation arises when a word-final McGregor, & Allen, 1992; Tallal & Piercy, coronal nasal or coronal stop is followed by a 1974), previous studies have rarely distinguished word-initial stop consonant with a different place between SLI children with and without dyslexia of articulation, for example: (McArthur et al., 2000). Of those that have, ten pens–te[m] pens, ten coins–te[˛] coins some studies report similar deficits—for example, good boy–goo[b] boy, good girl–goo[g] girl a deficit affecting the perception of rhythmic timing in both children with SLI only and those This process is termed regressive place assimilation with dyslexia only (Corriveau, Pasquini, & and is optional but widespread in connected Goswami, 2007; Goswami et al., 2002), while speech (Barry, 1985). others report differences; for example, children In cases such as these, it is not enough to know with dyslexia only but not those with SLI only that the phonological shape of the word ten is have a disadvantage for repeating consonant clus- /ten/, but that this phonological shape can be ters in unstressed compared to stressed syllables altered in a systematic way in connected speech. (Marshall & van der Lely, 2009). Despite the fact that these phonological changes In this study, we focus our attention on a can potentially disrupt lexical recognition, since hitherto relatively neglected aspect of phonology they can neutralize contrasts between phonemes, in children with dyslexia and SLI—namely, place they seem to matter little in everyday continuous assimilation. We argue that this approach allows speech. Native listeners are able to “compensate” us to investigate not only the quality of children’s for assimilation—that is, undo the change and phonological representations, but also their pho- thereby access the correct lexical form, due to nological grammar (which we define in due their implicit knowledge of which sound changes course). Furthermore, we address the issue of the are or are not allowed at word junctures in their overlap between the two disorders by comparing language. For example, English has place assimila- three groups: SLI + dyslexia, SLI only, and dys- tion but not voicing assimilation, whereas French lexia only. In the remainder of this introduction has voicing assimilation but not place assimilation. we discuss the phonological phenomenon of Hence football is often realized as foo[p]ball in place assimilation and why investigating assimila- English but as foo[d]ball in French. tion promises to shed further light on our under- One explanatory model of place assimilation is

Downloaded by [Harvard College] at 03:18 25 October 2011 standing of dyslexia and SLI. “phonological inference”, ﬁrst proposed by Knowing a word involves knowing the phono- Marslen-Wilson and colleagues (Marslen-Wilson, logical form of that word, a form that is idiosyn- Nix, & Gaskell, 1995). Phonological inference is a cratic in that it differs between languages. language-speciﬁc mechanism that undoes the effect Moreover, “the phonetic interpretation of a sen- of assimilation rules that apply during phonological tence makes reference not only to the phonological planning prior to production. This mechanism is shape of individual words but also to the phonolo- variously proposed to operate on the basis of some gical effects that result from combining these kind of rule-based “reverse” phonology (Gaskell & words in sequence” (Harris, 1994, p. 1). Marslen-Wilson, 1996) or through a statistically Importantly, a word’s phonological form can based recurrent connectionist model (Gaskell, change when it is produced next to other words. Hare, & Marslen-Wilson, 1995). Such phonological changes are not idiosyncratic: More recently, Darcy and her colleagues have They display regular patterning, in that they are proposed a model whereby compensation for

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 565 MARSHALL, RAMUS, VAN DER LELY

assimilation is driven by multiple cues: universal (i.e., the form that is actually uttered; Chomsky phonetic compensation for some coarticulation & Halle, 1968; Prince & Smolensky, 2004). cues, inverse phonological rules that are specific to Investigating assimilation therefore provides a the particular language in question, and lexical novel test of the accuracy of phonological rep- influences (Darcy, Ramus, Christophe, Kinzler, & resentations at the segmental level in children Dupoux, 2009). Darcy et al. ran a word detection with dyslexia and SLI: If phonological units are task, where participants pressed a button as soon poorly represented, then phonological as they detected a particular target word—for grammar—that is, the mapping process between example, brown. Participants were presented with a word’s underlying and surface form, in this case an assimilated form of the word, either in a at word junctures—would plausibly be affected. context in which assimilation was possible in To our knowledge, assimilation has not been English or French—for example, fa[p] puppy studied in children with SLI, but there have (place assimilation—possible in English), bla[g] been at least a couple of studies in individuals glove (voicing assimilation—possible in French)— with dyslexia. or a context in which it was impossible—for The effect of assimilatory context has been example, fa[p] squirrel, bla[g] rug. Darcy et al. tested in Dutch children with dyslexia aged 7–9 found that American English adults compensated years (Blomert et al., 2004). In a two-alternative more for place than for voicing assimilation, while forced-choice task, children were asked to report French adults showed the opposite pattern and whether they heard /m/ or /n/ in compound compensated more for voicing assimilation. words whose context for assimilation varied. However, they also found that the non-native They heard unassimilated and assimilated forms assimilation rule (i.e., voicing assimilation for of tuin in an appropriate context2 (tui[n]bank and English and place assimilation for French) tui[m]bank, “garden bench”), an inappropriate induced a small but significant compensation context (tui[n]stoel and tui[m]stoel, “garden effect, suggesting that both language-specific and chair”), or no context (tui[n] and tui[m]). The dys- language-independent mechanisms are at play. lexic children showed the same pattern of results as Lexical phonological representations must be their controls: Identification of the nasal was more abstract enough to encompass variability due to difficult in the appropriate context than in the voice, intonation, and linguistic context, but inappropriate or no context conditions, and there must also include enough phonetically relevant was a bias towards the canonical form tui[n]bank detail to discriminate near lexical neighbours and for both groups. However, the results of this to permit the child to learn about the various sys- study are not easily generalizable as they were tematic sources of variability in the sounds of obtained on a single pair of words. words. Therefore, phonological learning mechan- French adults with dyslexia were tested on

Downloaded by [Harvard College] at 03:18 25 October 2011 isms must be able to both abstract over and incor- Darcy et al.’s experimental stimuli (Szenkovits, porate phonetic details and information about Darma, Darcy, & Ramus, 2011). Adults with dys- words’ surrounding context (Fisher & Church, lexia compensated for voicing assimilations in 2001). Assimilatory processes are part of the lis- viable contexts to the same extent as did controls tener’s phonological grammar, where by and, again like the controls, rarely compensated “grammar” we mean the set of abstract rules or for place assimilation. Furthermore, in a pro- constraints that explain the mapping between the duction experiment, the dyslexic group produced underlying form of a word and the surface form voicing assimilation around 40% of the time in

2Here and in the rest of this paper, we use the terms “appropriate” or “viable” for a phonological context in which the target assimilation may occur, according to the phonology of the particular language under consideration (for example, here, in Dutch, a bilabial stop is an appropriate context for regressive place assimilation). Conversely, an inappropriate or unviable context is a context where such assimilation does not normally occur (here, a fricative is not an appropriate context for place assimilation in Dutch).

566 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

viable contexts but not in unviable contexts, as did We based our task on the English word- the controls. reporting task of Darcy et al. (2009), but adapted In summary then, it would appear that children it for children, making the sentences simpler and with dyslexia are able to acquire the phonological the procedure more child orientated. In their assimilation and compensation processes of their task and ours, target words are presented auditorily native language. That they are able to acquire this and are followed by a sentence containing the aspect of their phonological grammar suggests, target, and participants are asked to report when contrary to the received wisdom, that their phono- they hear the target word correctly pronounced. logical representations are detailed enough to rep- In the sentences, the target word surfaces either resent the relevant phonological contexts and the with a change of the final place feature or various forms of the segments undergoing assimila- without any change (baseline). The change tion. However, the experiments of Szenkovits et al. occurs either in a viable context or in an unviable (2011) were only carried out with well-compen- context. Participants press a button when they sated dyslexic adults and so need to be run on chil- think that the sentence contains the target word dren. Furthermore, assimilation has never been correctly pronounced. A yes response (button tested in English-speaking children with dyslexia, press) indicates that the word in the sentence is nor has it been tested in children with SLI. being treated as a token of the target, and a no Therefore we set out to investigate whether response (no button press) indicates that the English-speaking children with dyslexia and/or change altering the word blocks its interpretation SLI are impaired in their use of the place assimila- as a token of the target. tion rule, as compared with typically developing This design therefore permits us to obtain a children matched for age and different aspects of measure of the degree of tolerance for modifications reading and language ability. altering word forms and the degree to which this tolerance depends on phonological context in the way defined by the language’s phonological grammar. If a participant reports the changed word as canonical Method more often in viable than in unviable contexts, Background then that indicates that he or she has acquired sensi- In order to investigate participants’ implicit tivity to English place assimilation rules. For knowledge of assimilation, we consider context example, recognizing goo[b] as a good instance of effects: Occurrences of assimilation in the stimuli “good” when it is followed by “boy” but not when are either viable (i.e., surface in an appropriate it is followed by “friend” implies some knowledge context for assimilation) or unviable (the context of both the phonological features that may be assimi- is not normally a trigger for the modification). lated in English (place but not voicing) and of the Downloaded by [Harvard College] at 03:18 25 October 2011 There is a third condition (“no change”) in contexts in which they may be assimilated (plosives, which the target word surfaces without any but not fricatives). Thus such an effect can be change, in order to provide baseline performance explained neither by solely attending to the acous- for the ability to report the word within a sentence. tic/phonetic details of the target word, nor by These conditions are illustrated in Table 1. general compensation for coarticulation processes.

Table 1. Experimental conditions

Condition Example With carrier sentence

no change brow[n] roof My neighbour has an ugly brow[n] roof viable context brow[m] bell My neighbour has an ugly brow[m] bell unviable context brow[m] lamp My neighbour has an ugly brow[m] lamp

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 567 MARSHALL, RAMUS, VAN DER LELY

Participants abilities, we carried out some additional tests. Three clinical groups (SLI + dyslexia, SLI only, and We administered two nonstandardized language dyslexia only) and three control groups also partici- tests: the Test of Active and Passive Sentences pated in the study (two younger groups matched (TAPS; van der Lely, 1996)—a test of reversible for different aspects of language and literacy abilities active and passive sentence comprehension—and (LA1 and LA2, where “LA” stands for “language/ the Verb Agreement and Tense Test (VATT; literacy ability”), and one group matched for chrono- van der Lely, 2000)—a test of third-person agree- logical age (CA). The children in the clinical groups ment and past-tense marking on regular and irre- were recruited two years prior to the current study, to gular verbs. These two tests target language take part in a comprehensive investigation of phono- structures that are particularly impaired in SLI: logical abilities in SLI and/or dyslexia (Marshall, passive sentences and finite verb morphology. Harcourt-Brown, Ramus, & van der Lely, 2009; We calculated age-corrected z scores for the Marshall & van der Lely, 2009). Note that all partici- TAPS and the VATT on the basis of the control pants continued to be in special education during this data. In addition, we administered the Single time and were tested every six months as part of that Word Spelling and Comprehension subtests of investigation. The children in the clinical groups the WORD and the Nonword Reading subtest were between the ages of 8;00 and 12;11 years at of the Phonological Assessment Battery (PhAB; the time of recruitment and between 10;00 and Frederickson, Frith, & Reason, 1997). To gain a 15;00 years at the time of this particular study. The picture of participants’ phonological abilities, we following criteria were used to select children for used three other subtests of the Phonological the clinical groups: Assessment Battery—Rhyme, Spoonerisms, A minimum standard score of 80 on two tests of nonverbal cog- Rapid Naming (digits)—and also the Digit Span nition (Raven’s Standard Progressive Matrices, RPM, Raven, subtest (forwards and backwards) of the 1998; and the Block Design subtest from British Ability Wechsler Intelligence Scales for Children Scales–2, BAS, Elliott, 1996) and an average combined (WISC; Wechsler, 1992). These tests tap a range minimum score of 85 (i.e., –1 SD below the mean or higher); of different abilities that traditionally fall under no additional diagnoses of attention-deficit/hyperactivity disorder (ADHD), autistic spectrum disorder (ASD) or dyspraxia; the rubric of “phonology” in dyslexia research; a statement of special educational need and attendance at a the rhyme and spoonerisms tasks test phonological special school or unit for children with SLI or dyslexia. awareness and also require phonological represen- In addition, selection for SLI was based on the fol- tations to be held in phonological working lowing criteria: memory whilst they are being compared/manipulated; the rapid naming task requires rapid access A standard score of 78 or below (i.e., 7th percentile, z score of to lexical phonological representations; and the –1.5) on at least one of the following: Test for Reception of Grammar–2 (TROG; Bishop, 2003); British Picture digit span task requires phonological working Downloaded by [Harvard College] at 03:18 25 October 2011 Vocabulary Scales–2 (BPVS; Dunn, Dunn, Whetton, & memory. Burley, 1997); Sentence Repetition subtest of Clinical In order to test children’s phonological rep- Evaluation of Language Fundamentals–3 (CELF; Semel, resentations using a task that was unspeeded and Wiig, & Secord, 1995); Test of Word-Finding–2 (TWF; that had minimal working memory and metapho- German, 2000). nological demands, we created a picture–word Selection for the dyslexia group used the following matching task. This task allowed us to investigate criteria: whether children are able to distinguish between A standard score of 78 or below (i.e., 7th percentile, z score of two familiar phonological representations that –1.5) on the Single Word Reading subtest of the Wechsler differ only minimally. Stimuli were restricted to Objective Reading Dimensions (WORD; Wechsler, 1990), monosyllabic CVC, CV, or VC words (where C which comprises phonologically regular and irregular words. is consonant, V is vowel), presented in minimal In order to obtain a detailed profile of our par- pairs (see Appendix A for a full list). The pairs dif- ticipants’ language, literacy, and phonological fered in either initial or final consonant, and half

568 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

were presented with multitalker babble as back- + dyslexia group. For this round of testing, the ground noise (with a signal-to-noise ratio of numbers in each group were as follows: 28 SLI + 0 dB) in order to stress the child’s perceptual dyslexia, 10 SLI only, and 18 dyslexia only. system. The experiment was run on a laptop com- Children in the control groups had to have a puter that had a touch screen to record the partici- standard score of 85 or above on every language pants’ responses. In each trial, participants saw two and literacy task along with no history of speech pictures whose names differed only by one conso- or language delay or special educational needs. nant sound, and they heard the name of one of They were between 5;00 and 12;11 years of age those pictures. They had to touch the picture at the time of recruitment and 7;01–14;11 years that corresponded to the word that they heard. in this current round of testing. They were The results of all these tests are set out in divided into three age bands: LA1 (N ¼ 15), Tables 2 and 3. 7;01–8;06, mean 7;11; LA2 (N ¼ 16), 8;07– We found a substantial overlap between the SLI 10;00, mean 9;03; CA (N ¼ 30), 10;01–14;11, and dyslexia groups even though many of the chil- mean ¼ 11;10. The oldest group was a chrono- dren had an official diagnosis of only a single deficit. logical age-matched control group for the SLI Thus, many children fulfilled our criteria for both and dyslexic groups and therefore allowed us to SLI and dyslexia and so were assigned to the SLI investigate whether the phonological skills of

Table 2. Number of participants in each group, age of participants at time of experimental testing, and results of nonverbal and language tests

SLI + dyslexia SLI only Dyslexia only LA1 controls LA2 controls CA controls (N ¼ 28) (N ¼ 10) (N ¼ 18) (N ¼ 15) (N ¼ 16) (N ¼ 30)

Test Score M SD M SD M SD M SD M SD M SD

Age† 12.75 1.15 12.44 1.74 12.32 1.23 7.92 0.50 9.23 0.50 11.84 1.26 RPM raw 33.71 4.76 37.9 4.09 38.11 5.90 18.87 4.14 27.88 7.73 39.00 8.47 z –0.55 0.37 0.20 0.74 0.19 0.94 0.48 0.40 0.80 0.77 0.63 0.90 BAS‡ z –0.44 0.46 –0.18 0.57 0.22 0.77 0.77 1.23 0.47 1.03 0.70 0.96 TROG raw 10.75a 3.30 13.80b,c 2.30 16.22c,d 2.10 12.00a,b 3.66 14.81c,d 2.20 16.93d 1.91 z –1.64 0.91 –0.78 0.71 0.05 0.72 0.75 0.92 0.60 0.70 0.39 0.54 BPVS raw 78.14b 16.68 87.30b,c 14.60 98.89c,d 13.18 64.73a 10.63 85.19b 7.74 102.07d 13.12 z –1.28 0.76 –0.65 0.76 0.08 0.67 0.44 0.76 0.86 0.56 0.50 0.64 CELF raw 19.20a 8.96 23.95a,b 5.82 45.11c 12.65 30.47b 8.88 37.75b,c 9.27 52.50c 9.65 z –2.17 0.25 –2.00 0.42 –0.43 0.78 0.57 0.50 0.42 0.76 0.32 0.69 TWF raw 39.04a 10.24 50.90b 8.17 60.83b,c 7.98 39.80a 12.35 58.63b,c 6.28 64.73c 7.39 z –2.26 0.61 –1.25 0.61 –0.30 0.76 0.14 0.61 0.55 0.78 0.39 0.79

Downloaded by [Harvard College] at 03:18 25 October 2011 TAPS raw 26.14a 5.90 26.30a 5.52 32.83b 2.31 27.13a 5.77 30.06a,b 3.07 31.90b 3.60 z –1.37 1.28 –1.29 1.31 0.16 0.53 –0.12 1.19 0.21 0.65 0.05 0.78 VATT raw 17.18a 10.44 27.50b 6.87 34.28b,c 4.24 29.13b 6.72 33.81b,c 3.37 37.37c 2.22 z –4.09 1.95 –2.10 1.09 –0.67 0.81 –0.28 1.23 0.22 0.61 0.07 0.52

Note: For each test, raw scores that share a subscript do not differ significantly at the p ¼ .05 level on post hoc testing with Bonferroni correction. SLI ¼ specific language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. Key to tests: RPM: Raven’s Progressive Matrices (Raven, 1998). BAS: British Ability Scales, Block Design subtest (Elliott, 1996). TROG: Test of Reception of Grammar (Bishop, 2003). BPVS: British Picture Vocabulary Scales (Dunn, Dunn, Whetton, & Burley, 1997). CELF: Clinical Evaluation of Language Fundamentals, Sentence Repetition subtest (Semel, Wiig, & Secord, 1995). TWF: Test of Word Finding (German, 2000). TAPS: Test of Active and Passive Sentences (van der Lely, 1996). VATT: Verb Agreement and Tense Test (van der Lely, 2000). †Children were an average of 24 months younger than this at the time that the standardized language and literacy tests were administered. ‡We do not report raw scores for the BAS, because, depending on their age, children attempt a different number of items—the raw scores therefore vary in ways that do not reflect performance on this task.

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 569 ASAL AU,VNDRLELY DER VAN RAMUS, MARSHALL, 570 ONTV ERPYHLG,21,2 (7) 27 2010, NEUROPSYCHOLOGY, COGNITIVE

Table 3. Results of literacy and phonological tests

SLI + dyslexia SLI only Dyslexia only LA1 controls LA2 controls CA controls (N ¼ 28) (N ¼ 10) (N ¼ 18) (N ¼ 15) (N ¼ 16) (N ¼ 30)

Test Score M SD M SD M SD M SD M SD M SD

WORD Reading raw 18.39a 7.05 36.90b 9.55 20.00a 5.52 18.60a 9.66 34.63b 4.67 44.57c 5.86 z –2.07 0.46 –0.18 0.81 –1.83 0.33 1.20 0.95 1.39 0.88 0.96 0.73 Spelling raw 16.57a 4.32 27.40b 7.31 17.28a 3.04 15.20a 4.97 24.38b 4.83 33.17c 5.48 z –1.95 0.41 –0.43 0.54 –1.78 0.45 1.01 0.62 1.09 0.93 0.58 0.76 Comprehension raw 6.79a 4.86 16.30b,c 5.83 10.89a,b 6.35 8.40a 7.14 18.75c 3.87 27.30d 3.75 z –2.58 0.55 –1.10 0.72 –1.88 0.65 0.80 0.94 0.91 0.69 1.06 0.89

PhAB Nonword reading raw 6.12a 4.47 14.00b,c 4.88 9.39a,b 3.63 10.53b 4.87 14.75c 2.98 17.90c 2.92 z –1.03 0.55 –0.03 0.53 –0.56 0.49 0.77 0.34 0.67 0.34 1.06 0.75 Rhyme raw 8.21a 4.25 17.00c 3.77 12.28b 5.72 13.20b 5.02 17.50c 3.27 18.63c 1.69 z –1.56 0.49 –0.09 0.50 –0.82 1.15 0.82 0.33 0.88 0.57 0.48 0.60 Spoonerisms raw 6.89a 5.41 15.40b 7.43 11.11a,b 6.52 7.67a 2.58 15.00b 6.13 22.97c 4.78 z –1.18 0.54 –0.23 0.77 –0.63 0.60 0.38 0.15 0.78 0.59 0.97 0.65 Rapid naming raw 73.86a,b 23.67 54.80b,c 12.70 76.44a,b 23.74 94.07a 24.66 65.63b,c 15.87 47.37c 15.11 z –1.12 0.71 –0.19 0.69 –1.05 0.76 –0.24 0.43 0.53 0.79 0.78 0.95

WISC Digit span raw 9.43a 1.81 10.30a 2.21 10.94a 2.07 14.40b 3.81 13.69b 2.55 15.83b 3.01 z –1.57 0.48 –1.27 0.78 –1.04 0.50 1.13 1.15 0.58 0.85 0.49 0.91

Picture–word matching raw 91.46a,b 5.29 92.95a,b,c 3.85 94.94b,c 3.40 87.39a 9.36 91.54a,b,c 6.34 95.33c 5.07

Downloaded by [Harvard College] at 03:18 25 October 2011 October 25 at 03:18 College] by [Harvard Downloaded Note: For each test, raw scores that share a subscript do not differ signiﬁcantly at the p ¼ .05 level on post hoc testing with Bonferroni correction. SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. Key to tests: WORD: Wechsler Objective Reading Dimensions (Wechsler, 1990). PhAB: Phonological Assessment Battery (Frederickson, Frith, & Reason, 1997). WISC: Wechsler Intelligence Scales for Children (Wechsler, 1992). PLACE ASSIMILATION IN DYSLEXIA AND SLI

children with SLI and dyslexia fall below age Characteristics of control groups expectations. However, as described in the next Younger controls with comparable linguistic and section, the two younger control groups allowed literacy abilities are widely used in research with us to investigate more closely the relationship children who have dyslexia and/or SLI. If per- between phonology, literacy, and language abilities formance on a particular task falls below chrono- such as word and sentence comprehension. logical age expectations, this method of matching There were significant group differences on all allows researchers to determine whether the the language, literacy, and phonology tests, as lower performance is still in line with general measured by a series of one-way analyses of var- language and literacy abilities, or whether it falls iance (ANOVAs) on raw scores (p , .001 for below even those expectations. Typically in each). Differing subscripts in Tables 2 and 3 indi- research with children who have dyslexia and cate significant differences on post hoc testing SLI, performance on phonological tasks is at or with Bonferroni correction. At this point in the below that expected from general language and lit- text, we highlight the results for the measures eracy abilities (e.g., Corriveau et al., 2007; Gallon that were not used as admission criteria to the et al., 2007; Joanisse et al., 2000). different clinical groups. For the two nonstandar- In line with a previous study (Marshall et al., dized tests, tapping syntax and morphosyntax— 2009), we carried out post hoc analyses to deter- the TAPS and the VATT—the two SLI groups, mine which control groups provided the best but not the dyslexia-only group, score significantly matches for each clinical group for language and below chronological age expectations (see Table literacy abilities—namely, for the comprehension 2). For the PhAB nonword reading task, this of a mixed range of sentences (TROG), single pattern is, not surprisingly, reversed—this time word comprehension (BPVS), and reading abil- it is the two dyslexia groups, but not the SLI ities (WORD reading). These analyses reveal group, that fall below chronological age expec- that the LA1 group provides the best match to tations. For the phonological tasks—namely, the the SLI + dyslexia group for sentence compre- other three subtests of the PhAB (Rhyme, hension and to both the dyslexia groups for Spoonerisms, and Rapid Naming) and the single word reading. The LA2 group provides WISC Digit Span—the two dyslexia groups also the best match to the SLI + dyslexia group for fall below chronological age expectations, as vocabulary comprehension and to the SLI-only would be expected from many previous studies group for all three tasks. In contrast, the CA of dyslexia. The SLI-only group also shows weak- group provides the best match to the dyslexia- nesses in phonology, but only falls below chrono- only group with respect to sentence and vocabulary logical age expectations for two tasks: comprehension. These matches are set out in Spoonerisms and Digit Span. Table 4 for clarity. Downloaded by [Harvard College] at 03:18 25 October 2011

Table 4. Group matches

Language/literacy measure LA1 controls LA2 controls CA controls

Sentence comprehension: TROG SLI + dyslexia SLI only Dyslexia only Receptive vocabulary: BPVS SLI + dyslexia Dyslexia only SLI only Single word reading: WORD SLI + dyslexia SLI only Dyslexia only

Note: SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. Key to tests: TROG: Test of Reception of Grammar (Bishop, 2003). BPVS: British Picture Vocabulary Scales (Dunn, Dunn, Whetton, & Burley, 1997). WORD: Wechsler Objective Reading Dimensions (Wechsler, 1990).

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 571 MARSHALL, RAMUS, VAN DER LELY

Stimuli chea[b] rooms;targetfine, sentence containing final A total of 16 target adjectives were chosen, 8 score). Crucially, these training sentences did not ending in /n/,4in/d/, and 4 in /t/ (see contain any cases of place assimilation in either Appendix B for a full list). Each of the target viable or unviable contexts, so the feedback could adjectives was associated with a triplet of context not affect responses on test sentences. nouns. Each adjective in a triplet corresponded Sentences were recorded by a female speaker to one of the experimental conditions—no (the first author), and the target words by a male change, viable change, and unviable change. For speaker, both with standard southern British the viable-change condition, the noun’s initial English accents. Post hoc, and as part of the exper- consonant was a stop consonant that could iment on adult participants described below, we trigger place assimilation of the adjective-final verified that the target words had indeed been pro- consonant (e.g., brown bell–brow[m] bell; fat duced as intended. puppy–fa[p] puppy). For the unviable-change and no-change conditions, the noun’s initial consonant Procedure was a nonstop consonant and could not trigger The experiment was programmed in Visual Basic. place assimilation of the adjective-final consonant. A wizard appeared on the left hand side of the However, in the stimuli that participants heard, screen and said the target word; 500 ms later a adjectives in the viable and unviable change con- girl appeared on the right-hand side of the ditions had undergone assimilation (incorrectly, screen and said a sentence. Participants were of course, in the case of the unviable change con- requested to press a button when they thought dition; e.g., brown lamp–brow[m] lamp; fat squir- they heard the girl say the wizard’s (i.e., target) rel–fa[p] squirrel). word in her sentence exactly as he had said it Three sentence frames were constructed for and to refrain from pressing the button otherwise. each of the target items. A sentence frame con- This instruction—together with the specific train- sisted of a sentence beginning and sentence ing—was given so that they paid attention to the ending, where each of the three adjective–noun actual form of the words. Participants were told combinations could be inserted to create a plaus- to respond as quickly as possible, without waiting ible sentence (e.g., My neighbour put a brow[n] until the end of the sentence. They had up to roof above his door; My neighbour put a brow[m] one second following the end of the sentence to bell above his door; My neighbour put a brow[m] make their reply if they needed it. lamp above his door). Presentation of each trial was controlled by the For the purposes of counterbalancing, we made experimenter. Instructions were presented orally. three experimental lists. In each list, all three con- During the training phase, feedback was provided ditions were present for each item, but in different in the shape of a green smiley face for a correct

Downloaded by [Harvard College] at 03:18 25 October 2011 sentence frames. The sentence frames were rotated response and a red sad face for an incorrect across the three lists, so that across the experimen- response. There was no feedback in the test tal lists all three conditions appeared in all three phase. Each of the three lists was pseudorando- sentence frames. mized, and, within each list, each participant Participants first participated in a training trial received items in the same order. The entire pro- with 18 training sentences using a different set of cedure lasted approximately 12 minutes. adjectives, where they received feedback as to whether their answer was correct or incorrect. Predictions Modifications involved voicing, manner, and place Correct understanding of this task should be contrasts at the ends of words, or deletion of a reflected by the participants’ ability to report word-final syllable or alteration of a rime, in order target words in the no-change condition and to to drive the participants’ attention to the precise reject them in the unviable-context condition. form of words (e.g., target cheap, sentence containing The performance of these two conditions serves

572 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

as a baseline to evaluate the responses in the viable- . “Impaired acquisition of phonological rules” context condition. If participants fully compensate hypothesis: Alternatively, they might have a for the phonological rule, they should (erro- more specific deficit in the acquisition of pho- neously) report the target word to the same nological rules, possibly as a particular manifes- extent as in the no-change condition, despite the tation of their grammatical impairment in the fact that the target underwent the same featural phonological domain. This would predict change as in the unviable condition. If there is again that they would be less sensitive to the no compensation for the phonological rule, par- phonological context. Two opposite predictions ticipants should respond as in the unviable- might follow: They might have an overall ten- context condition—that is, reject (correctly) the dency not to compensate for assimilation, and changed word as a target. Given previous studies, so would not report target words in either we did not predict full compensation for assimila- viable nor unviable contexts, or they might tion. In the study by Darcy et al. (2009), adult have a tendency to overcompensate, reporting American English listeners compensated for target words in both viable and unviable place assimilation 46% of the time in viable con- contexts. texts. Therefore, the difference in acceptance of . “Intact phonological representations” hypothesis: the changed word between the viable and unviable Yet another hypothesis would be that they do conditions can be seen as an index of language- not have degraded phonological representations specific phonological compensation. (Ramus & Szenkovits, 2008; Soroli et al., 2010), We predict a developmental progression within or at least not at the segmental level (van der control children, whereby the magnitude of the Lely & Marshall, 2011). This would predict difference between acceptance in the viable and no difference with the control children. unviable conditions increases with age. Given pre- . “General task difficulties” hypothesis: Finally, they vious results for voicing assimilation in French and might have more general difficulties performing place assimilation in Dutch, we might expect the the task. Indeed the word-reporting task is quite dyslexia-only group to show the same pattern as complex, so its performance could be affected by the controls. However, for all three clinical various cognitive limitations, including: (a) groups (SLI + dyslexia, SLI only, and dyslexia underspecification of phonological lexical only), there are several possible predictions forms; (b) poor metalinguistic abilities; (c) depending on one’s hypothesis about the nature poor vocabulary (inducing less familiarity with of their deficit, and the predictions need not be target words); (d) poor verbal short-term the same for the three groups. memory (recognizing words within sentences requires verbal short-term memory; Jacquemot, . “Degraded phonological representations” hypothesis: Dupoux, Decouche, & Bachoud-Levi, 2006);

Downloaded by [Harvard College] at 03:18 25 October 2011 If children with dyslexia and/or SLI have (e) poor inhibition (word reporting needs to be degraded phonological representations at the inhibited in the unviable context). All these segmental level (which is predicted in particular task performance factors would predict less by auditory theories; Tallal et al., 1998; Tomblin difference between the two baseline conditions & Pandich, 1999), then one would predict that (no change and unviable context). Note that they would show fewer consistent responses this hypothesis is not incompatible with any of overall (because of less precise representation the previous three, so that the results might of target phonemes) and less sensitivity to reﬂect a superimposition of several of the pre- phonological context—that is, less difference dictions proposed here. We speciﬁcally test between viable and unviable context conditions whether our measures of vocabulary size, (because of less precise representations of pho- verbal short-term memory, metaphonological nological contexts). abilities, and simple (CVC) phonological

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 573 MARSHALL, RAMUS, VAN DER LELY

representations at the word level predict per- conditions, F2(2, 47) ¼ 50.441, p , .001. Paired- formance on the assimilation task. sample t tests again showed that the differences between all conditions were highly significant: Checking the adequacy of the stimuli viable versus unviable, t(15) ¼ –5.500, p , .001; Before we ran the experiment with children, we no change versus unviable, t(15) ¼ 10.829, p , tested the materials on adults in two ways. First, .001; and no change versus viable, t(15) ¼ 6.483, we ran a replication of Darcy et al.’s (2009) p , .001. Although detection rates in the viable word-reporting study, using the procedures out- and unviable conditions were about 10% higher lined previously. The aim was to check that the than those reported by Darcy et al. (2009; they stimuli to be used with the children produced found approximately 50% detection for the viable the same pattern of results with adults as Darcy and 15% detection for the unviable conditions), et al.’s stimuli had. We then ran a control task, the basic pattern of results was the same. whereby participants were asked to categorize the To assess whether the critical items’ final con- target words that had been extracted from their sonants had been produced as prescribed by the carrier sentences. Here the aim was to check that experimental conditions, after adults had com- the stimuli used in the viable and unviable con- pleted the word-reporting task they undertook a ditions were acoustically comparable. control task. All target words were extracted A group of 11 adults participated. All were from their carrier sentences and were presented speakers of British English recruited from in isolation in a forced-choice categorization University College London and were between task. The motivation for this task was to check the ages of 18 and 38 years. Their word-reporting that the difference we observed between the data are presented in Table 5. viable and unviable conditions in the main task A repeated measures ANOVA by subjects was not simply due to differences in the target revealed a significant difference between the words: Rather, it was due to differences in the adults’ performance in the different conditions of context in which those words were presented. the word-reporting task (i.e., the no change, Words were presented auditorily and were fol- viable, and unviable), F1(2, 20) ¼ 100.83, p , lowed by a 3,000-ms silence, during which partici- .001. Paired-sample t tests showed that the differ- pants had to circle the consonant they heard on the ences between all conditions were highly signifi- response sheet. Participants were always given cant: viable versus unviable, t(10) ¼ –7.692, p , the choice between the original consonant and .001; no change versus unviable, t(10) ¼ 15.539, the assimilated one. For example, for the word p , .001; and no change versus viable, t(10) ¼ “brown” the choice was between “n” and “m”. 6.145, p , .001. A one-way ANOVA by items However, they also had a free cell in which to also showed highly significant differences between place an alternative consonant if that is what Downloaded by [Harvard College] at 03:18 25 October 2011

Table 5. Responses: Percentage reporting of target word

SLI + Adult dyslexia SLI only Dyslexia only LA1 LA2 CA

MSDMSDMSDMSDMSDMSDMSD

No change 97.16 4.30 93.95 8.05 94.33 4.62 88.54 15.79 91.52 9.37 92.97 7.53 93.07 6.35 Viable 61.93 18.64 87.45 12.47 80.63 23.28 81.94 12.84 77.17 16.12 76.17 20.44 72.20 20.63 Unviable 25.00 13.11 80.62 16.64 73.75 22.20 58.68 23.59 66.07 20.02 55.86 24.53 43.68 21.63

Note: Illustrative example. SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. Target word to be reported: brown No-change condition: brow[n] roof Viable condition: brow[m] bell Unviable condition: brow[m] lamp

574 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

Table 6. Results for the forced-choice categorization of extracted We explored the interaction by first carrying out a words series of one-way ANOVAs to detect significant group differences within each condition. For the % final consonants perceived as changed no-change condition there was no effect of group, F(5, 106) ¼ 0.92, p ¼ .472. The group effect for Condition MSDthe viable condition just missed significance, ¼ ¼ No change (e.g., brow[n]) 2.86 2.17 F(5, 106) 2.24, p .056. For the unviable con- Viable change (e.g., brow[m]) 92.38 6.23 dition, however, the effect of group was significant, Unviable change (e.g., brow[m]) 87.86 7.65 F(5, 106) ¼ 9.21, p , .001. Post hoc testing (Bonferroni corrected) revealed that the SLI + Adult group only. Note: dyslexia group reported the target word signifi- they heard. The entire procedure lasted about 15 cantly more often on the unviable condition than ¼ ¼ minutes. The results are shown in Table 6. did the dyslexia-only (p .017), LA2 (p .006) The rates of perception of the final consonant as and CA (p , .001) groups, while the SLI-only “changed” are more accurate than those reported in group reported the target word significantly more ¼ Darcy et al’s (2009) experiment (which were 23%, often than did the CA group, p .003. No other 74%, and 78% for the no-change, viable, and unvi- group differences approached significance. Both able conditions, respectively). A Wilcoxon signed SLI groups therefore perform below chronological ranks t test revealed that the difference between age expectations with respect to word reporting in + the viable and unviable conditions just missed sig- the unviable condition, with the SLI dyslexia nificance, Z(9) ¼ –1.834, p ¼ .067. Therefore, group performing below their vocabulary- there was a trend for target words from viable con- matched controls (i.e., the LA2 group). Both texts to be pronounced more assimilated than groups are, however, performing in line with their words from unviable contexts. This difference sentence comprehension and word reading controls + could only have reduced the likelihood of report- (i.e., LA1 for the SLI dyslexia group, and LA2 ing the target words in viable contexts in the for the SLI-only group). main experiment and therefore makes our esti- Next, we tested the factor “condition” within each group. For all groups this was significant: SLI + mation of compensation for assimilation rather 2 dyslexia, F(2, 48) ¼ 11.56, p , .001, hp ¼ .325; conservative. 2 SLI only, F(2, 18) ¼ 5.24, p ¼ .016, hp ¼ .368; 2 dyslexia only, F(2, 34) ¼ 33.08, p , .001, hp ¼ Results 2 .661; LA1, F(2, 26) ¼ 15.88, p , .001, hp ¼ + 2 Two children with SLI dyslexia refused to com- .550; LA2 (2, 30) ¼ 31.10, p , .001, hp ¼ .675; 2 plete the task, and a technical failure resulted in CA (2, 56) ¼ 82.55, p , .001, hp ¼ .747.

Downloaded by [Harvard College] at 03:18 25 October 2011 data from one child in the LA1 control group The high rates of reporting in the unviable con- not being saved. We removed responses where dition (see Table 5) suggest that there are biases in the reaction time was negative or below 150 ms. word reporting. Further, these rates seem to vary This amounted to 2.84% of the data from the across groups, with the SLI + dyslexia group three clinical groups and 2.41% of the data from showing the highest levels and the CA group the the three control groups. Results are shown in lowest. Therefore, the best way to evaluate the sig- Table 5. nificance of compensation effects, and to compare A 3 (condition: no change, viable, unviable) × 6 them across groups, is to carry out a signal detec- (group) ANOVA revealed a significant interaction tion analysis. between condition and group, F(10, 212) ¼ 6.71, 2 p , .001, hp ¼ .240, and main effects of condition, Signal detection analysis 2 F(2, 212) ¼ 132.35, p , .001, hp ¼ .555, and We use signal detection analysis to separate word 2 group, F(5, 106) ¼ 5.26, p , .001, hp ¼ .199. reporting from response bias, using a classification

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 575 MARSHALL, RAMUS, VAN DER LELY

model with three stimuli (no change, assimilated better than chance performance in this task (see ′ in viable context, assimilated in unviable context) Figure 1). However, word-reporting d differed assumed to be represented along a single dimen- significantly between the groups of children, F(5, sion (similarity to the target word) and classified 111) ¼ 8.35, p , .001. Post hoc testing into two categories (present or absent; Macmillan (Bonferroni corrected) revealed that the SLI + & Creelman, 2005). dyslexia group differed significantly from the We first recoded our reporting rates into LA2 and CA groups (p ¼ .011 and p , .001, hits (baseline hit rate ¼ % word reporting in respectively) and that the LA1 group differed sig- the no-change condition, and compensation hit nificantly from the CA group (p ¼ .040). No rate ¼ % word reporting in the viable condition) other differences reached significance. and false alarms (false-alarm rate ¼ % word All groups have significantly negative bias reporting in the unviable condition) and computed values (one-sample t tests, all p values ,.001), as ′ two d values: illustrated in Figure 2, reflecting a general ten- ′ dency to respond “word” by default whenever in . Word-reporting d ¼ z (baseline hit rate) – doubt. This is understandable given that in the z(false-alarm rate), where z is the inverse of unviable condition, participants had to refrain the normal distribution function. This measure from pressing the button despite hearing a string indexes the sensitivity of overall word reporting that was minimally different from the target in the task. ′ word. Furthermore, the unique identification . Compensation d ¼ z (compensation hit rate) – point of the word was likely to have been z (false-alarm rate) indexes the sensitivity of reached prior to the end of the word, which assimilated word reporting specifically in the could also trigger the participants pressing the viable context. As such, it indexes the degree ′ button. Bias is correlated with word-reporting d to which compensation for assimilation is (R ¼ .598, p , .001). Thus, the more difficult specific to the contexts defined by English pho- the children find the task, the more likely they nological grammar. are to adopt a liberal bias. . We also calculated the overall bias for producing ′ Finally, compensation d reflects the ability to a “word” response, where bias ¼ z (false-alarm compensate for place assimilation by reporting rate). words in viable as opposed to unviable contexts. ′ These results are shown in Table 7 and Figures Compensation d is significantly greater than 0 1 and 2. for all groups—SLI + dyslexia, t(24) ¼ 2.590, p ′ Word-reporting d indexes the general ability ¼ .016; dyslexia only, t(17) ¼ 5.377, p , .001; to perform the word-reporting task. All groups LA1, t(13) ¼ 2.924, p ¼ .012; LA2, t(15) ¼ ′ showed a d significantly greater than 0 (one- 5.533, p , .001; CA, t(28) ¼ 9.529, p , .001— Downloaded by [Harvard College] at 03:18 25 October 2011 sample t tests, all p values , .05), reflecting their except for the SLI-only group, t(9) ¼ 2.005,

′ ′ Table 7. Word-reporting d , compensation d , and bias

SLI + dyslexia SLI only Dyslexia only LA1 LA2 CA Adults

MSDMSDMSDMSDMSDMSDMSD

′ Word-reporting d 0.55 0.62 0.77 0.94 0.68 1.29 0.94 0.68 1.29 0.62 1.68 0.73 2.48 0.48 ′ Compensation d 0.27 0.53 0.29 0.35 0.45 0.64 0.35 0.45 0.64 0.46 0.88 0.50 1.11 0.54 Bias –1.27 0.42 –1.16 0.96 –0.84 0.43 –0.96 0.43 –0.84 0.48 –0.65 0.34 –0.49 0.18

Note: SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age.

576 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

′ ′ Figure 1. Word-reporting d and compensation d . SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. dys ¼ dyslexia. Downloaded by [Harvard College] at 03:18 25 October 2011

Figure 2. Boxplots of bias values. SLI ¼ speciﬁc language impairment. LA ¼ language/literacy ability. CA ¼ chronological age. dys ¼ dyslexia.

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 577 MARSHALL, RAMUS, VAN DER LELY

′ ′ Table 8. Correlation matrix between compensation d , word-reporting d , and all their potential predictors, with age partialled out

PhAB PhAB Digit Picture–word ′ Word-reporting d RPM BPVS rhyme spoonerisms span matching

′ Compensation d R .585 .216 .344 .311 .349 .240 .280 p ,.001 .024 ,.001 .001 ,.001 .012 .003 ′ Word-reporting d R .362 .557 .519 .454 .422 .393 p ,.001 ,.001 ,.001 ,.001 ,.001 ,.001 RPM R .358 .348 .377 .099 .329 p ,.001 ,.001 ,.001 .306 ,.001 BPVS R .455 .420 .389 .369 p ,.001 ,.001 ,.001 ,.001 PhAB rhyme R .690 .483 .306 p ,.001 ,.001 .001 PhAB spoonerisms R .519 .388 p ,.001 ,.001 Digit span R .325 p .001

Note: Key to tests: RPM: Raven’s Progressive Matrices (Raven, 1998). BPVS: British Picture Vocabulary Scales (Dunn, Dunn, Whetton, & Burley, 1997). PhAB: Phonological Assessment Battery (Frederickson, Frith, & Reason, 1997).

p ¼ .080. Although the effect is not significant for sentence? Rerunning the 3 (condition: no the SLI-only group, compared to the SLI + dys- change, viable, unviable) × 6 (group) ANOVA ′ ′ lexia group compensation d is actually larger for with word-reporting d as a covariate results in a ′ the SLI-only group and the variance smaller. significant effect of word-reporting d , F(1, 112) Therefore we assume that it is the low sample ¼ 34.08, p , .001, but group no longer emerges size of the SLI-only group that decreased the stat- as a significant factor, F(5, 112) ¼ 1.52, p ¼ istical power of the test. Hence there is good evi- .189. We therefore conclude that, once word- dence that all the groups do compensate for reporting abilities are controlled for, there are no assimilation to some extent. However, there are significant differences in the magnitude of the significant group differences, F(5, 112) ¼ 5.51, p compensation effect between groups. , .001. Post hoc testing (Bonferroni corrected) In order to assess to what extent average results for group reveals that the SLI + dyslexia group reported above are representative of each group, ′ has a significantly lower compensation d than boxplots are shown in Supplementary Figures 1 the dyslexia-only group, p ¼ .037, and than the and 2. They show in particular that at least 75%

Downloaded by [Harvard College] at 03:18 25 October 2011 CA group, p , .001. The SLI-only and LA1 of the children in the SLI groups and 100% of ′ ′ groups also have a lower compensation d than children in the dyslexia-only group had d values the CA group, p ¼ .033 and p ¼ .030, greater or equal than 0, which suggests that the respectively. average compensation abilities reported are However, children’s possibility of reporting representative. words differently in the viable and unviable con- ′ texts (compensation d ) is intrinsically limited by Predictors of word-reporting abilities their capacity to perform the word detection task In order to test the hypothesis that verbal short- ′ correctly (word-reporting d ). The question then term memory, metaphonological skills, and/or ′ is: Do group differences in compensation d phonological representations at the word level reﬂect differences speciﬁcally in the ability to com- affect word-reporting rates in our cohort of SLI, pensate for assimilation, or are they simply due to dyslexic, and control children, we multiple linear ′ differences in how the groups report words in a regressions of word-reporting d with all the

578 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

relevant predictors that we had in our data set. then vocabulary entered last. This was done ′ Because word-reporting d indexes absolute per- using either all putative predictors or only the formance, we used only raw (rather than age- significant ones as revealed by the previous analy- standardized) scores of the relevant variables. sis, and with or without the group factor. In all Table 8 shows the partial correlation matrix cases, vocabulary explained significant additional ′ ′ between compensation d , word-reporting d , variance (at least 5%) and became the leading and all their potential predictors, with age par- predictor of the complete regression model tialled out. (semipartial R . .23). We carried out multiple stepwise linear These results suggest that the tolerance for ′ regressions with word-reporting d as the dependent minimal deviations from canonical word forms variable and with the following regressors: age, the decreases primarily with vocabulary growth, and Raven’s Progressive Matrices, BPVS (receptive secondarily with metaphonological abilities and vocabulary), rhyme awareness and spoonerisms with the precision of phonological representations. (both indexing metaphonological abilities), digit Taken together, these three factors seem sufficient span (indexing verbal working memory), and the to account for the observed group differences, and task testing the precision of segmental phonological thus for SLI children’s limited word-reporting representations in simple CVC words: picture– abilities. word matching (all raw scores). Finally, we ran a similar stepwise regression ′ We found that rhyme awareness entered first with compensation d as the dependent variable into the model, F(1, 108) ¼ 34.7, p , .001, and including the very same predictors plus ′ adjusted R2 ¼ .24, then vocabulary, F(1, 107) ¼ word-reporting d We found that only word- ′ 17.6, p , .001, additional R2 ¼ .11, then age, reporting d entered the model, F(1, 108) ¼ 60, F(1, 106) ¼ 4.7, p ¼ .03, additional R2 ¼ .03, p , .001, explaining 35.1% of the variance. and finally picture–word matching, F(1, 105) ¼ Furthermore, the group factor did not explain 4.4, p ¼ .04, additional R2 ¼ .03. The total additional variance, F(1, 107) ¼ 0.61. This confirms amount of variance explained was 37.9% (adjusted that compensation abilities are limited mostly by R2). Entering the group factor3 in a last step did word-reporting abilities. not significantly increase the amount of variance explained, F(1, 104) ¼ 1.7, p ¼ .20, suggesting that these predictors adequately account for the Discussion and conclusions observed group differences in word-reporting This study set out to test whether children with abilities. In the final model, vocabulary shows SLI and/or dyslexia compensate for lawful pho- the greatest contribution to word-reporting nological variation during lexical access, by ′ ¼ d (semipartial R .31), then rhyme awareness investigating their ability to compensate for Downloaded by [Harvard College] at 03:18 25 October 2011 ¼ (semipartial R .24), then age (semipartial place assimilation in a word-reporting task. We ¼ 4 R –.20) and picture–word matching (semi- argued that this approach allows us to investi- ¼ partial R .16). gate whether phonological representations at We tested the robustness of vocabulary as ′ the segmental level are impaired in these the main predictor of word-reporting d by con- groups and in addition allows us to investigate structing various regression models with all pre- one aspect of their phonological grammar (by dictors but vocabulary entered simultaneously, which we mean the set of abstract rules or

3For the purpose of this analysis, the group factor is a four-level nominal variable coding for each of the three pathological groups and the control group. ′ 4The negative partial correlation between age and word-reporting d can be explained by the fact that the raw scores in the model carry most of the age-related variance. However, absolute performance levels off with age, hence the additional negative contribution of age.

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 579 MARSHALL, RAMUS, VAN DER LELY

′ constraints that map a surface form to its under- cases their compensation d should be lower than lying form). normal, which was not the case. Our findings Our results are as follows: (a) Signal detection therefore replicate and extend the results obtained analyses reveal that children in all three clinical for Dutch children by Blomert et al. (2004) and for groups (SLI + dyslexia, SLI only, and dyslexia French adults by Szenkovits et al. (2011). only) compensate for place assimilation; (b) differ- The groups of children with SLI + dyslexia ences in the magnitude of compensation effects and SLI only differ from the dyslexia-only group across groups are primarily due to differences in in that they are more biased towards giving a word-reporting abilities; and (c) vocabulary level “word” response: They are more likely to accept is the main predictor of word-reporting abilities. the assimilated form of a word, regardless of Our central finding is that children with SLI whether the word was presented in a viable or an and/or dyslexia do indeed compensate for place unviable context. Yet, despite this strong response assimilation, in the specific context-dependent bias, they do show a significant sensitivity to the manner prescribed by English phonological phonological context in which assimilations were grammar. The group of children with dyslexia presented, compensating more in viable than in only do not differ from their age-matched peers unviable contexts: This suggests that they are in their performance on the word-reporting task. able to learn the context-specific place assimilation In contrast, the two SLI groups do not perform rule. The results are not easily reconciled with a age appropriately on the task. Yet the performance degraded phonological representations hypothesis, of the two SLI groups can be explained by task or with a hypothesis that they might have an effects: They are more liberal in their acceptance impairment in acquiring phonological rules. ′ of alternative phonological forms of words, but The numerically low compensation d values do not differ from the control and dyslexia-only observed in SLI and dyslexia groups may seem to groups in anything specific to compensation for conflict with this conclusion. However, it should ′ place assimilation. be noted that compensation d cannot get very The age-appropriate performance of children high given that participants are not expected to with dyslexia in the assimilation task stands in compensate all the time. Indeed, even adult par- marked contrast to their performance on other ticipants compensate for place assimilation in phonological tasks—namely, rhyming, spooner- only about 60% of the trials, which is consistent isms, rapid naming, and digit span, where they with the previous study by Darcy et al. (2009). It ′ performed significantly below their chronologi- follows that their compensation d is only around ′ cal-age controls (and, in the case of the digit 1.1, less than half their word-reporting d .Not span task, more poorly than their reading-age con- surprisingly, children show a similar pattern, but ′ trols). Our results suggest that children with dys- since they have a lower word-reporting d to start ′ Downloaded by [Harvard College] at 03:18 25 October 2011 lexia are able to acquire implicit knowledge of with, their compensation d is correspondingly this aspect of the phonological grammar of their lower. native language. Furthermore, they suggest that The main source of concern in the present study children with dyslexia have relatively accurate rep- is word-reporting abilities, which are more liberal resentations of the phonological features that may than we would have hoped for. It is undeniable be assimilated, as well as of the phonological con- that the task was complex and was perhaps not texts in which assimilation may take place. Such fully understood or correctly performed by all chil- results are difficult to reconcile with the hypothesis dren. The main consequence is to increase the that dyslexic children have degraded phonological noise in our data and to limit our ability to representations: That hypothesis would predict observe specific compensation effects. Future either less compensation for assimilation, or a replications of this study should certainly aim to more generalized pattern of compensation (less further reduce the complexity of this task, dependent on the phonological context). In both improve the instructions and training, and

580 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

therefore enhance children’s overall performance. That phonological representations have some Nevertheless, these limitations played against our predictive role in word-reporting abilities is not ability to detect compensation effects. It is there- surprising: A child who has difficulty perceiving fore remarkable that in spite of these difficulties, the difference between those minimal phonetic we observed a significantly positive compensation differences (as between “browm” and “brown”) is ′ d even in the children with the most severe dis- likely to match both forms to the lexical item. orders, which supports the conclusion that they However, this does not seem to be the major have acquired a sensitivity to the phonological problem in these SLI children, most of whom contexts for place assimilation in English. have high levels of performance in picture–word Even so, children’s liberal acceptance of matching. On the other hand, metaphonological alternative phonological forms of words, particu- skills also predicted word-reporting abilities, pre- larly amongst the two SLI groups, requires an sumably because judging whether a word is a explanation. Of course, some degree of tolerance good phonetic match (vs. a mispronunciation) for minor deviations from the canonical word taps metaphonology, an area where children with form is expected, since speech errors are SLI and dyslexia have difficulties. common. Listeners must therefore have a Once word-reporting abilities are controlled, general bias towards automatically assimilating group effects in compensation abilities disap- minimally deviant pseudowords to the nearest pear—that is, lose statistical significance. This lexical item. This is indeed reflected in the bias may be seen as a null result, and one might worry measure of our adult participants. Nevertheless, that our statistical power to detect group differences ′ this tolerance is limited, and listeners do readily in compensation d while controlling for word- ′ notice many speech errors. Furthermore, there is reporting d was limited. Indeed, the observed a clear developmental trend—that is, a reduction power for this test was 51.6%, so the conclusion of this tolerance with vocabulary growth, as that compensation abilities are really the same shown in our data by the effect of raw vocabulary across all groups remains tentative. A further ′ scores on word-reporting d . Regression analyses caveat with respect to our interpretation is that showed that vocabulary size was the main devel- the participants with SLI and dyslexia were tested opmental predictor of the ability to reject mispro- when they were between 10 and 15 years of age, nunciations of words, but that metaphonological and so we can say nothing about their development skills, and to a minor extent phonological rep- of compensation for assimilation at a younger age. resentations, also played a part. The poorer per- Despite these caveats, we conclude that our formance of the two SLI groups on the results add to a growing body of evidence that assimilation task is therefore explained by the phonological representations are not necessarily demands of the task. impaired in dyslexia. The less accurate perform-

Downloaded by [Harvard College] at 03:18 25 October 2011 A possible explanation for vocabulary size as a ance of the children with SLI in the task reduced predictor of word-reporting ability is that while the range of performance within which their com- young children have a relatively small lexicon, pensation for assimilation could be observed. phonological neighbourhoods are sparse, and Nevertheless, within that limited range, they did there are few competitors for word recognition. show a sensitivity to the phonological contexts in Any word form in the neighbourhood of a lexical which target words were embedded and compen- item can be assimilated to this item, unless the sated more for assimilation in viable than in unvi- semantic context dictates that these must be two able contexts. Thus, it seems that children with different words. On the other hand, as the SLI, as well as those with dyslexia, are able to lexicon grows, phonological neighbourhoods learn this aspect of phonological grammar. become denser, and there is a need to consider The growing body of evidence that phonologi- ﬁner phonetic details to access the correct lexical cal representations are not necessarily impaired items. in dyslexia challenges models attributing the

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 581 MARSHALL, RAMUS, VAN DER LELY

phonological deficit in dyslexia to degraded or nature of the phonological representation itself, altered phonological representations. It is more are needed. Finally, a developmental perspective commensurate with proposals that the deficit is is of course essential, so that a picture of phonolo- in accessing phonological representations (Blomert gical development and its relationship to language et al., 2004; Dickie, 2008; Ramus & Szenkovits, and literacy over the lifespan can be constructed. 2008; Soroli et al., 2010). According to this latter proposal, it is when phonological tasks Manuscript received 30 June 2010 place heavy demands on short-term memory, con- Revised manuscript received 28 April 2011 scious awareness, and time constraints that indi- Revised manuscript accepted 4 May 2011 First published online 28 June 2011 viduals with dyslexia perform poorly. Another, compatible, theoretical proposal attributes the phonological deficit to short- and medium-term storage of phonological representations (the “anchoring” deficit; Ahissar, 2007). Importantly, REFERENCES our proposal does not challenge the existence of a phonological deficit in dyslexia, but it makes Adlard, A., & Hazan, V. (1998). Speech perception in the claim that the phonological deficit does not children with specific reading difficulties (dyslexia). lie in the quality of the phonological represen- Quarterly Journal of Experimental Psychology, 51A, tations themselves. 153–177 Ahissar, M. (2007). Dyslexia and the anchoring-deficit Finally, with respect to models of the overlap hypothesis. Trends in Cognitive Sciences, 11, 458–465. between dyslexia and SLI, this is the third study Barry, M. C. (1985). A palatographic study of connected investigating phonology in this same cohort of speech processes. Cambridge Papers in Phonetics and participants with dyslexia only, SLI only, and Experimental Linguistics, 4, 1–16. SLI + dyslexia. On a nonword repetition task Bishop, D. V. M. (2003). Test for reception of grammar. that manipulated the position of clusters in non- London, UK: Harcourt Assessment, Psychological words, all groups performed poorly compared to Corporation. even language-matched children, and yet there Bishop, D. V. M., & Snowling, M. J. (2004). were qualitative differences between the SLI- Developmental dyslexia and specific language only group and the two groups with dyslexia impairment: Same or different? Psychological (Marshall & van der Lely, 2009). In a separate Bulletin, 130, 858–888. Blomert, L., Mitterer, H., & Paffen, C. (2004). In test of prosodic skills, using the Profiling search of the auditory, phonetic, and/or phonologi- Elements of Prosodic Systems–Child Version cal problems in dyslexia: Context effects in speech (Peppe´ & McCann, 2003), no differences were perception. Journal of Speech, Language, and Hearing found between the dyslexia and SLI groups, and Research, 47, 1030–1047. Downloaded by [Harvard College] at 03:18 25 October 2011 few children had difficulties with the tasks, Bortolini, U., & Leonard, L. (2000). Phonology and which involved same/different judgements of, children with specific language impairment: Status and imitation of, prosodic forms (Marshall et al., of structural constraints in two languages. Journal of 2009). Taken together, the results of the three Communication Disorders, 33, 131–150. studies suggest that phonological impairments Bradley, L., & Bryant, P. E. (1983). Categorising are indeed where dyslexia and SLI overlap, but sounds and learning to read: A causal connection. that the phonological impairments in the two Nature, 301, 419–421. Castles, A., & Coltheart, M. (2004). Is there a causal disorders are not necessarily identical, nor are link from phonological awareness to success in learn- they necessarily present in every domain of pho- ing to read? Cognition, 91, 77–111. nology. Further studies of the type presented in Catts, H., Adlof, S., Hogan, T., & Weismer, S. E. here, using tasks that move beyond the traditional (2005). Are specific language impairment and dys- phonological awareness, rapid naming, and short- lexia distinct disorders? Journal of Speech, Language term memory tasks, and instead focusing on the and Hearing Research, 48, 1378–1396.

582 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

Chomsky, N., & Halle, M. (1968). The sound pattern of envelope onsets and developmental dyslexia: A new English. New York, NY: Harper & Row. hypothesis. Proceedings of the National Academy of Corriveau, K., Pasquini, E., & Goswami, U. (2007). Sciences, 99, 10911–10916. Basic auditory processing skills and specific language Guttorm, T. K., Leppa¨nen, P.H. T., Richardson, U., & impairment: A new look at an old hypothesis. Journal Lyytinen, H. (2001). Event-related potentials and of Speech, Language, and Hearing Research, 50, consonant differentiation in newborns with familial 647–666. risk for dyslexia. Journal of Learning Disabilities, 34, Darcy, I., Ramus, F., Christophe, A., Kinzler, K., & 534–544. Dupoux, E. (2009). Phonological knowledge in com- Hari, R., & Renvall, H. (2001). Impaired processing of pensation for native and non-native assimilation. In rapid stimulus sequences in dyslexia. Trends in F. Ku¨gler, C. Fe´ry, & R. Vijverv. d. (Eds.), Variation Cognitive Sciences, 5, 525–532. and gradience in phonetics and phonology Harris, J. (1994). English sound structure. Oxford, UK: (pp. 265–309). Berlin, Germany: Mouton De Gruyter. Blackwell. Dickie, C. (2008). Exploring the nature of the phonological Jacquemot, C., Dupoux, E., Decouche, O., & deficit in dyslexia: Are phonological representations Bachoud-Levi, A. C. (2006). Misperception in sen- impaired? Unpublished doctoral dissertation, tences but not in words: Speech perception and the University of Edinburgh, Edinburgh, UK phonological buffer. Cognitive Neuropsychology, 23, Dunn, L. M., Dunn, L. M., Whetton, C., & Burley, J. 949–971. (1997). British Picture Vocabulary Scales (2nd ed.). Joanisse, M., Manis, F., Keating, P., & Seidenberg, M. Windsor, UK: NFER Nelson. S. (2000). Language deficits in dyslexic children: Elliott, C. D. (1996). British Ability Scales (2nd ed.). Speech perception, phonology and morphology. Windsor, UK: NFER Nelson. Journal of Experimental Child Psychology, 77, 30–60. Fisher, C., & Church, B. (2001). Implicit memory Kamhi, A., & Catts, H. (1986). Toward an understand- support for language acquisition. In J. Weissenborn ing of developmental language and reading disorders. & B. Hoehle (Eds.), Approaches to bootstrapping: Journal of Speech, Language, and Hearing Research, 51, Phonological, lexical, syntactic and neurophysiological 337–347. aspects of early language acquisition (Vol. 1). Leonard, L. (1998). Children with specific language Amsterdam, The Netherlands: John Benjamins. impairment. Cambridge, MA: MIT Press. Frederickson, N., Frith, U., & Reason, R. (1997). Leonard, L., McGregor, K., & Allen, G. (1992). Phonological Assessment Battery. Windsor, UK: Grammatical morphology and speech perception in NFER Nelson. children with specific language impairment. Journal Gallon,N.,Harris,J.,&vanderLely,H.K.J. of Speech, Language, and Hearing Research, 35, (2007). Non-word repetition: An investigation 1076–1085. of phonological complexity in children with Lundberg, I., & Høien, T. (1989). Phonemic deficits: A grammatical-SLI. Clinical Linguistics and core symptom of developmental dyslexia? Irish Phonetics, 21, 435–455. Journal of Psychology, 10, 579–592. Gaskell, M. G., Hare, M., & Marslen-Wilson, W. Lyytinen, H., Aro, M., Eklund, K., Erskine, J.,

Downloaded by [Harvard College] at 03:18 25 October 2011 (1995). A connectionist model of phonological rep- Guttorm, T., Laakso, M. L., et al. (2004). The resentation in speech perception. Cognitive Science, development of children at familial risk for dyslexia: 19, 407–439. Birth to early school age. Annals of Dyslexia, 54, Gaskell, M. G., & Marslen-Wilson, W. (1996). 184–220. Phonological variation and inference in lexical Macmillan, N. A., & Creelman, C. D. (2005). Detection access. Journal of Experimental Psychology: Human theory: A user’s guide (2nd ed.). Mahwah, NJ: Perception and Performance, 22, 144–158. Lawrence Erlbaum Associates. German, D. J. (2000). Test of Word Finding (2nd ed.). Marshall, C. R., Harcourt-Brown, S., Ramus, F., & van Austin, TX: Pro-ed. der Lely, H. K. J. (2009). The link between prosody Goswami, U. (2011). A temporal sampling framework and language skills in children with SLI and/or dys- for developmental dyslexia. Trends in Cognitive lexia. International Journal of Language & Sciences, 15, 3–10. Communication Disorders, 44, 466–468. Goswami, U., Thomson, J., Richardson, U., Stainthorp, Marshall, C. R., & van der Lely, H. K. J. (2009). Effects R., Hughes, D., S., Rosen, et al. (2002). Amplitude of word position and stress on onset cluster

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 583 MARSHALL, RAMUS, VAN DER LELY

production: Data from typical development, SLI and Serniclaes, W., Van Heghe, S., Mousty, P., Carre, R., & dyslexia. Language, 85, 39–57. Sprenger-Charolles, L. (2004). Allophonic mode Marslen-Wilson, W., Nix, A., & Gaskell, M. G. (1995). of speech perception in dyslexia. Journal of Phonological variation in lexical access: Abstractness, Experimental Child Psychology, 87, 336–361. inference and English place assimilation. Language Snowling, M. J. (2000). Dyslexia (2nd ed.). Oxford, UK: and Cognitive Processes, 10, 285–308. Blackwell. Mason, K., Rowley, K., Marshall, C. R., Atkinson, J. R., Soroli, E., Szenkovits, G., & Ramus, F. (2010). Exploring Herman, R., Woll, B., et al. (2010). Identifying SLI dyslexics’ phonological deficit: III. Foreign speech per- in deaf children acquiring British Sign Language: ception and production. Dyslexia, 16, 318–340. Implications for theory and practice. British Journal Sperling, A. J., Lu, Z. L., Manis, F. R., & Seidenberg, of Developmental Psychology, 28, 33–49. M. S. (2005). Deficits in perceptual noise exclusion McArthur, G. M., Hogben, J. H., Edwards, V. T., in developmental dyslexia. Nature Neuroscience, 8, Heath, S. M., & Mengler, E. D. (2000). On the 862–863. “specifics” of specific reading disability and specific Stein, J. F., & Fowler, M. S. (1981). Visual dyslexia. language impairment. Journal of Child Psychology Trends in Neurosciences, 1, 77–80. and Psychiatry, 41, 869–874. Stein, J. F., & Walsh, V. (1997). To see; but not to read: Messaoud-Galusi, S., & Marshall, C. R. (2010). The magnocellular theory of dyslexia. Trends in Introduction to this special issue. The overlap Neurosciences, 20, 147–151. between SLI and dyslexia: The role of phonology. Szenkovits, G., Darma, Q., Darcy, I., & Ramus, Scientific Studies of Reading, 14, 1–7. F. (2011). Exploring dyslexics’ phonological deficit II: Mody, M., Studdert-Kennedy, M., & Brady, S. (1997). Phonological grammar. Manuscript submitted for Speech perception deficits in poor readers: Auditory publication. processing or phonological encoding. Journal of Tallal, P. (2003). Language learning disabilities: Experimental Child Psychology, 64, 199–231. Integrating research approaches. Current Directions Molfese, D. L. (2000). Predicting dyslexia at 8 years of in Psychological Science, 12, 206–211. age using neonatal brain responses. Brain and Tallal, P., Merzenich, M., Miller, S., & Jenkins, I. Language, 72, 238–245. (1998). Language learning impairments: Nicolson, R. I., & Fawcett, A. J. (2007). Procedural Integrating basic science, technology and remedia- learning difficulties: Reuniting the developmental tion. Experimental Brain Research, 123, 210–219. disorders? Trends in Neurosciences, 30, 135–141. Tallal, P., & Piercy, M. (1974). Developmental aphasia: Pennington, B. F., & Bishop, D.V. M. (2009). Relations Rate of auditory processing and selective impairment among speech, language and reading disorders. of consonant perception. Neuropsychologia, 12, Annual Review of Psychology, 60, 283–306. 83–93. Peppe´, S., & McCann, J. (2003). Assessing intonation Tomblin, J. B., & Pandich, J. (1999). Lessons from chil- and prosody in children with atypical language devel- dren with specific language impairment. Trends in opment: The PEPS-C test and the revised version. Cognitive Sciences, 3, 283–285. Clinical Linguistics and Phonetics, 17, 345–354. Valdois, S., Bosse, M. L., & Tainturier, M. J.

Downloaded by [Harvard College] at 03:18 25 October 2011 Prince, A., & Smolensky, P. (2004). Optimality theory: (2004). The cognitive deficits responsible for Constraint interaction in generative grammar. developmental dyslexia: Review of evidence for a Oxford, UK: Wiley-Blackwell. selective visual attentional disorder. Dyslexia, 10, Ramus, F., & Szenkovits, G. (2008). What phonological 339–363. deficit? Quarterly Journal of Experimental Psychology, van der Lely, H. K. J. (1996). Test of Active and Passive 61, 129–141. Sentences Available from http://www.dldcn.com Raven, J. C. (1998). Raven’s Progressive Matrices. van der Lely, H. K. J. (2000). Verb Agreement and Tense Oxford, UK: Oxford Psychologists Press. Test. Available from http://www.dldcn.com Scarborough, H. S. (1990). Very early language deficits van der Lely, H.K. J., & Marshall, C. R. (2010). in dyslexic children. Child Development, 61, Assessing component language deficits in the early 1728–1743. detection of reading difficulty risk. Journal of Semel, E., Wiig, E., & Secord, W. (1995). Clinical Learning Disabilities, 43, 357–368. evaluation of language fundamentals. San Antonio, van der Lely, H. K. J., & Marshall, C. R. (2011). TX: Psychological Corporation. Grammatical-specific language impairment: A

584 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) PLACE ASSIMILATION IN DYSLEXIA AND SLI

window onto domain-speciﬁcity. In J. Guendouzi, Wagner, R. K., & Torgesen, J. K. (1987). The nature of F. Loncke, & M. J. Williams (Eds.), The handbook phonological processing and its causal role in the of psycholinguistic and cognitive processes: Perspectives acquisition of reading skills. Psychological Bulletin, in communication disorders (pp. 401–418). London, 101, 192–212. UK: Taylor & Francis. Wechsler,D.(1990).Wechsler Objective Reading Dimensions. Vidyasagar, T. R., & Pammer, K. (2010). Dyslexia: A London, UK: The Psychological Corporation. deﬁcit in visuo-spatial attention, not in phonological Wechsler, D. (1992). Wechsler Intelligence Scale for Children. processing. Trends in Cognitive Sciences, 14, 57–63. London, UK: The Psychological Corporation.

APPENDIX A

Items for the picture–word matching task

Word-initial contrast Word-ﬁnal contrast

Manipulation Target Distractor Target Distractor

Noisy Voice pear bear eyes ice bull pull cart card down town peg peck sip zip back bag Place cap tap mud mug pick tick lake late tall call rat wrap fox socks wait wake Manner ﬁll pill late lace sew toe beep beef pan fan rose road top shop put push No noise Voice coat goat bad bat tie die seat seed fan van log lock buy pie dice dies Place date gate neck net tight kite pack pat

Downloaded by [Harvard College] at 03:18 25 October 2011 pea tea knot knock sat fat harp heart Manner shoe two hiss hit full pull card cars tail sail mash mat pin ﬁn toes toad

Note that our participants have nonrhotic accents, therefore for them words such as “heart” and “harp” do not contain ﬁnal clusters.

COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7) 585 MARSHALL, RAMUS, VAN DER LELY

APPENDIX B

Adjectives for the assimilation experiment

Table B1. Target words, their various forms, and their contexts

Target Final consonant Unchanged form Changed form No-change context Unviable context Viable context

lean n lean leam shadow silhouette body plain n plain plaim landscape churches buildings fun n fun fung day drive game thin n thin thim leaf sheet book clean n clean cleam fork spoon pan brown n brown browm roof lamp bell own n own owm choice life plan green n green greeng salad lemon grapefruit wild d wild wilg swan snake goose sad d sad sab song role book red d red reg mattress mirror garment cold d cold colb meat dish beer fat t fat fap monkey squirrel puppy short t short shorp ﬁght match play sweet t sweet sweek melon liqueur cocktail smart t smart smark mac suit coat Downloaded by [Harvard College] at 03:18 25 October 2011

586 COGNITIVE NEUROPSYCHOLOGY, 2010, 27 (7)